AI Infrastructure Capex Boom Faces Monetization Crisis

The AI infrastructure capex boom represents one of the most concentrated capital spending cycles in corporate history, with the five major hyperscalers (Microsoft, Alphabet, Amazon, Meta, Oracle) collectively committing $660-690 billion in 2026 capex — nearly doubling 2025 levels. Approximately 75% (~$450B) targets AI-specific infrastructure. NVIDIA dominates with ~85% GPU market share, posting $57B in Q3 FY2026 revenue (up 62% YoY) and guiding Q4 to $65B. Data center REITs like Equinix ($10.1-10.2B 2026 revenue guidance, +42% bookings growth) and Digital Realty (record 2025 results, $1.4B backlog) are clear beneficiaries of the physical infrastructure buildout. However, a significant monetization gap threatens the sustainability of this cycle. Sequoia Capital calculates that AI companies need $600 billion in annual revenue to justify current infrastructure spending, while actual AI-related revenue sits near $100 billion — a 6x gap. Enterprise AI adoption has reached 78%, but only 26% of companies report tangible value from AI investments. An MIT study found 95% of generative AI pilot programs fail to achieve business value. The capex-to-revenue disconnect is the central risk: hyperscalers now spend 45-57% of revenue on capex (resembling utilities, not tech companies), and aggregate capex after buybacks and dividends exceeds projected cash flows, necessitating record debt issuance. The credit implications are substantial. Hyperscalers issued $121 billion in bonds in 2025 (4x the five-year average), with Meta's $30B deal and Oracle's $25B offering in early 2026 headlining record issuance. Credit spreads have widened 2-5bps across investment grade and 20bps in tech specifically. Oracle's 5-year CDS has tripled since September, and Barclays warns Oracle may exhaust cash by November 2026. Morgan Stanley projects the sector may need $1.5 trillion in new debt over coming years. A structural vulnerability exists: GPU hardware depreciates in 3-5 years versus decades for traditional infrastructure, creating a perpetual reinvestment cycle. The macro backdrop adds complexity. The Fed holds rates at 3.5-3.75% with internal divisions on the path forward, the 10-year Treasury yields ~4.08%, and inflation remains above target at 2.4%. Goldman Sachs projects AI will boost productivity meaningfully starting in 2027, not 2026, suggesting the payoff from current spending is still over the horizon. Power grid constraints are binding — global data center electricity consumption is projected to double between 2022 and 2026, with some regions already dedicating 15-26% of electricity to data centers. This creates both a bottleneck for buildout and an investment opportunity in utilities and energy infrastructure. The thesis is fundamentally a timing question: if monetization materializes by late 2026/2027, current valuations look reasonable; if not, the sector faces a significant correction risk.